Weakly supervised object recognition with convolutional neural networks

By Maxime Oquab, L. Eon Bottou, Ivan Laptev, Josef Sivic. 2014

94 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Webly Supervised Learning of Convolutional Networks (2015, methodology): “The outline of our approach is shown in Figure 2.”
Fully Convolutional Multi-Class Multiple Instance Learning (2014, background): “Oquab et al. (2014) improve image classiﬁcation by inferring latent object location, but do not evaluate the localization.”
Return of the Devil in the Details: Delving Deep into Convolutional Nets (2014, methodology): “While the CNN-based methods of [16], [17] achieve better performance on VOC (86.3% and 90.3% respectively), they were trained using extended ILSVRC datasets, enriched with additional categories semantically close to the ones in VOC.”
Object Detectors Emerge in Deep Scene CNNs (2014, background): “Another set of recent works (Oquab et al., 2014; Bergamo et al., 2014) demonstrate the ability of deep networks trained on object classiﬁcation to do localization without bounding box supervision.”
Weakly-and Semi-Supervised Learning of a Deep Convolutional Network for Semantic Image Segmentation (2015, methodology): “While this approach has worked well for image classification tasks [28, 29], it is less suited for segmentation as it does not promote full object coverage: The DCNN becomes tuned to focus on the most distinctive object parts (e.”
From image-level to pixel-level labeling with Convolutional Networks (2014, methodology): “In a recent yet unpublished work [18], the authors adapt an Imagenet-trained CNN to the Pascal VOC classification task.”
Contextual Action Recognition with R*CNN (2015, methodology): “[24] modify the CNN architecture [17], which divides the image into equal sized regions and combines their scores via a final max pooling layer to classify the whole image.”
HCP: A Flexible CNN Framework for Multi-Label Image Classification (2014, neutral): “7 PRE-1512* [31] 94.”
Self-taught object localization with deep networks (2014, methodology): “mages. This framework enables scalable training of object detectors at a much reduced annotation cost. The idea of exploiting class label annotations to generate object locations was also explored in [24], however their work focused on the classication task, without providing any quantitative result”
Attention-based Deep Multiple Instance Learning (2018, neutral): “Similarly, in computer vision the idea of MIL combined with deep learning sig-niﬁcantly improves ﬁnal accuracy (Oquab et al., 2014).”
Deep Multi-patch Aggregation Network for Image Style, Aesthetics, and Quality Estimation (2015, methodology): “In [27, 20, 15], researchers computed ImageNet features (i.”
Multilabel Image Classification With Regional Latent Semantic Dependencies (2016, methodology): “art methods, including the EdgeBox [53], PRE-1000C* [56] and HCP-VGG [8].”
Deep Learning for Semantic Part Segmentation with High-Level Guidance (2015, methodology): “We employ it in a fully-convolutional manner , turning it into a dense feature extractor for semantic image segmentation, as in Sermanet et al. (2014); Oquab et al. (2014); Long et al. (2014), treating the last fully-connected layers of the DCNN as 1 × 1 spatial convolution kernels.”
Modeling local and global deformations in Deep Learning: Epitomic convolution, Multiple Instance Learning, and sliding window detection (2015, methodology): “We note that a similar object localization idea is also proposed in [29] for unsupervised object discovery, but without the maximization over scales.”
Weakly Supervised Deep Detection Networks (2015, methodology): “[25] employ a pre-trained CNN to compute a mid-level image representation for images of PASCAL VOC.”
A Multi-scale Multiple Instance Video Description Network (2015, background): “Each score in the output score map corresponds to one receptive field in the input image.”
Exploit Bounding Box Annotations for Multi-Label Object Recognition (2015, methodology): “To address the problems with global CNN representations, following the recent works [19, 26, 18, 12], we incorporate local information via extracting object proposals using general object detection techniques such as selective search [25] for multi-label object recognition.”
DTFD-MIL: Double-Tier Feature Distillation Multiple Instance Learning for Histopathology Whole Slide Image Classification (2022, background): “There exist some successful attempts to solve the MIL problem in various computer vision tasks [19,27,28,30,32].”